]>
Commit | Line | Data |
---|---|---|
749cf76c CD |
1 | /* |
2 | * Copyright (C) 2012 - Virtual Open Systems and Columbia University | |
3 | * Author: Christoffer Dall <c.dall@virtualopensystems.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License, version 2, as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. | |
17 | */ | |
342cd0ab CD |
18 | |
19 | #include <linux/mman.h> | |
20 | #include <linux/kvm_host.h> | |
21 | #include <linux/io.h> | |
45e96ea6 | 22 | #include <trace/events/kvm.h> |
342cd0ab | 23 | #include <asm/pgalloc.h> |
94f8e641 | 24 | #include <asm/cacheflush.h> |
342cd0ab CD |
25 | #include <asm/kvm_arm.h> |
26 | #include <asm/kvm_mmu.h> | |
45e96ea6 | 27 | #include <asm/kvm_mmio.h> |
d5d8184d | 28 | #include <asm/kvm_asm.h> |
94f8e641 | 29 | #include <asm/kvm_emulate.h> |
d5d8184d CD |
30 | |
31 | #include "trace.h" | |
342cd0ab CD |
32 | |
33 | extern char __hyp_idmap_text_start[], __hyp_idmap_text_end[]; | |
34 | ||
5a677ce0 | 35 | static pgd_t *boot_hyp_pgd; |
2fb41059 | 36 | static pgd_t *hyp_pgd; |
342cd0ab CD |
37 | static DEFINE_MUTEX(kvm_hyp_pgd_mutex); |
38 | ||
5a677ce0 MZ |
39 | static void *init_bounce_page; |
40 | static unsigned long hyp_idmap_start; | |
41 | static unsigned long hyp_idmap_end; | |
42 | static phys_addr_t hyp_idmap_vector; | |
43 | ||
48762767 | 44 | static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa) |
d5d8184d | 45 | { |
48762767 | 46 | kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa); |
d5d8184d CD |
47 | } |
48 | ||
d5d8184d CD |
49 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
50 | int min, int max) | |
51 | { | |
52 | void *page; | |
53 | ||
54 | BUG_ON(max > KVM_NR_MEM_OBJS); | |
55 | if (cache->nobjs >= min) | |
56 | return 0; | |
57 | while (cache->nobjs < max) { | |
58 | page = (void *)__get_free_page(PGALLOC_GFP); | |
59 | if (!page) | |
60 | return -ENOMEM; | |
61 | cache->objects[cache->nobjs++] = page; | |
62 | } | |
63 | return 0; | |
64 | } | |
65 | ||
66 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
67 | { | |
68 | while (mc->nobjs) | |
69 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
70 | } | |
71 | ||
72 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
73 | { | |
74 | void *p; | |
75 | ||
76 | BUG_ON(!mc || !mc->nobjs); | |
77 | p = mc->objects[--mc->nobjs]; | |
78 | return p; | |
79 | } | |
80 | ||
4f728276 | 81 | static void clear_pud_entry(pud_t *pud) |
342cd0ab | 82 | { |
4f728276 MZ |
83 | pmd_t *pmd_table = pmd_offset(pud, 0); |
84 | pud_clear(pud); | |
85 | pmd_free(NULL, pmd_table); | |
86 | put_page(virt_to_page(pud)); | |
87 | } | |
342cd0ab | 88 | |
4f728276 MZ |
89 | static void clear_pmd_entry(pmd_t *pmd) |
90 | { | |
91 | pte_t *pte_table = pte_offset_kernel(pmd, 0); | |
92 | pmd_clear(pmd); | |
93 | pte_free_kernel(NULL, pte_table); | |
94 | put_page(virt_to_page(pmd)); | |
95 | } | |
96 | ||
97 | static bool pmd_empty(pmd_t *pmd) | |
98 | { | |
99 | struct page *pmd_page = virt_to_page(pmd); | |
100 | return page_count(pmd_page) == 1; | |
101 | } | |
102 | ||
103 | static void clear_pte_entry(pte_t *pte) | |
104 | { | |
105 | if (pte_present(*pte)) { | |
106 | kvm_set_pte(pte, __pte(0)); | |
107 | put_page(virt_to_page(pte)); | |
342cd0ab CD |
108 | } |
109 | } | |
110 | ||
4f728276 MZ |
111 | static bool pte_empty(pte_t *pte) |
112 | { | |
113 | struct page *pte_page = virt_to_page(pte); | |
114 | return page_count(pte_page) == 1; | |
115 | } | |
116 | ||
117 | static void unmap_range(pgd_t *pgdp, unsigned long long start, u64 size) | |
000d3996 MZ |
118 | { |
119 | pgd_t *pgd; | |
120 | pud_t *pud; | |
121 | pmd_t *pmd; | |
4f728276 MZ |
122 | pte_t *pte; |
123 | unsigned long long addr = start, end = start + size; | |
124 | u64 range; | |
000d3996 | 125 | |
4f728276 MZ |
126 | while (addr < end) { |
127 | pgd = pgdp + pgd_index(addr); | |
128 | pud = pud_offset(pgd, addr); | |
129 | if (pud_none(*pud)) { | |
130 | addr += PUD_SIZE; | |
131 | continue; | |
132 | } | |
000d3996 | 133 | |
4f728276 MZ |
134 | pmd = pmd_offset(pud, addr); |
135 | if (pmd_none(*pmd)) { | |
136 | addr += PMD_SIZE; | |
137 | continue; | |
138 | } | |
000d3996 | 139 | |
4f728276 MZ |
140 | pte = pte_offset_kernel(pmd, addr); |
141 | clear_pte_entry(pte); | |
142 | range = PAGE_SIZE; | |
143 | ||
144 | /* If we emptied the pte, walk back up the ladder */ | |
145 | if (pte_empty(pte)) { | |
146 | clear_pmd_entry(pmd); | |
147 | range = PMD_SIZE; | |
148 | if (pmd_empty(pmd)) { | |
149 | clear_pud_entry(pud); | |
150 | range = PUD_SIZE; | |
151 | } | |
152 | } | |
153 | ||
154 | addr += range; | |
155 | } | |
000d3996 MZ |
156 | } |
157 | ||
d157f4a5 MZ |
158 | /** |
159 | * free_boot_hyp_pgd - free HYP boot page tables | |
160 | * | |
161 | * Free the HYP boot page tables. The bounce page is also freed. | |
162 | */ | |
163 | void free_boot_hyp_pgd(void) | |
164 | { | |
165 | mutex_lock(&kvm_hyp_pgd_mutex); | |
166 | ||
167 | if (boot_hyp_pgd) { | |
168 | unmap_range(boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE); | |
169 | unmap_range(boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); | |
170 | kfree(boot_hyp_pgd); | |
171 | boot_hyp_pgd = NULL; | |
172 | } | |
173 | ||
174 | if (hyp_pgd) | |
175 | unmap_range(hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); | |
176 | ||
177 | kfree(init_bounce_page); | |
178 | init_bounce_page = NULL; | |
179 | ||
180 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
181 | } | |
182 | ||
342cd0ab | 183 | /** |
4f728276 | 184 | * free_hyp_pgds - free Hyp-mode page tables |
342cd0ab | 185 | * |
5a677ce0 MZ |
186 | * Assumes hyp_pgd is a page table used strictly in Hyp-mode and |
187 | * therefore contains either mappings in the kernel memory area (above | |
188 | * PAGE_OFFSET), or device mappings in the vmalloc range (from | |
189 | * VMALLOC_START to VMALLOC_END). | |
190 | * | |
191 | * boot_hyp_pgd should only map two pages for the init code. | |
342cd0ab | 192 | */ |
4f728276 | 193 | void free_hyp_pgds(void) |
342cd0ab | 194 | { |
342cd0ab CD |
195 | unsigned long addr; |
196 | ||
d157f4a5 | 197 | free_boot_hyp_pgd(); |
4f728276 | 198 | |
d157f4a5 | 199 | mutex_lock(&kvm_hyp_pgd_mutex); |
5a677ce0 | 200 | |
4f728276 MZ |
201 | if (hyp_pgd) { |
202 | for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) | |
203 | unmap_range(hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); | |
204 | for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) | |
205 | unmap_range(hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); | |
206 | kfree(hyp_pgd); | |
d157f4a5 | 207 | hyp_pgd = NULL; |
4f728276 MZ |
208 | } |
209 | ||
342cd0ab CD |
210 | mutex_unlock(&kvm_hyp_pgd_mutex); |
211 | } | |
212 | ||
213 | static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start, | |
6060df84 MZ |
214 | unsigned long end, unsigned long pfn, |
215 | pgprot_t prot) | |
342cd0ab CD |
216 | { |
217 | pte_t *pte; | |
218 | unsigned long addr; | |
342cd0ab | 219 | |
3562c76d MZ |
220 | addr = start; |
221 | do { | |
6060df84 MZ |
222 | pte = pte_offset_kernel(pmd, addr); |
223 | kvm_set_pte(pte, pfn_pte(pfn, prot)); | |
4f728276 | 224 | get_page(virt_to_page(pte)); |
5a677ce0 | 225 | kvm_flush_dcache_to_poc(pte, sizeof(*pte)); |
6060df84 | 226 | pfn++; |
3562c76d | 227 | } while (addr += PAGE_SIZE, addr != end); |
342cd0ab CD |
228 | } |
229 | ||
230 | static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start, | |
6060df84 MZ |
231 | unsigned long end, unsigned long pfn, |
232 | pgprot_t prot) | |
342cd0ab CD |
233 | { |
234 | pmd_t *pmd; | |
235 | pte_t *pte; | |
236 | unsigned long addr, next; | |
237 | ||
3562c76d MZ |
238 | addr = start; |
239 | do { | |
6060df84 | 240 | pmd = pmd_offset(pud, addr); |
342cd0ab CD |
241 | |
242 | BUG_ON(pmd_sect(*pmd)); | |
243 | ||
244 | if (pmd_none(*pmd)) { | |
6060df84 | 245 | pte = pte_alloc_one_kernel(NULL, addr); |
342cd0ab CD |
246 | if (!pte) { |
247 | kvm_err("Cannot allocate Hyp pte\n"); | |
248 | return -ENOMEM; | |
249 | } | |
250 | pmd_populate_kernel(NULL, pmd, pte); | |
4f728276 | 251 | get_page(virt_to_page(pmd)); |
5a677ce0 | 252 | kvm_flush_dcache_to_poc(pmd, sizeof(*pmd)); |
342cd0ab CD |
253 | } |
254 | ||
255 | next = pmd_addr_end(addr, end); | |
256 | ||
6060df84 MZ |
257 | create_hyp_pte_mappings(pmd, addr, next, pfn, prot); |
258 | pfn += (next - addr) >> PAGE_SHIFT; | |
3562c76d | 259 | } while (addr = next, addr != end); |
342cd0ab CD |
260 | |
261 | return 0; | |
262 | } | |
263 | ||
6060df84 MZ |
264 | static int __create_hyp_mappings(pgd_t *pgdp, |
265 | unsigned long start, unsigned long end, | |
266 | unsigned long pfn, pgprot_t prot) | |
342cd0ab | 267 | { |
342cd0ab CD |
268 | pgd_t *pgd; |
269 | pud_t *pud; | |
270 | pmd_t *pmd; | |
271 | unsigned long addr, next; | |
272 | int err = 0; | |
273 | ||
342cd0ab | 274 | mutex_lock(&kvm_hyp_pgd_mutex); |
3562c76d MZ |
275 | addr = start & PAGE_MASK; |
276 | end = PAGE_ALIGN(end); | |
277 | do { | |
6060df84 MZ |
278 | pgd = pgdp + pgd_index(addr); |
279 | pud = pud_offset(pgd, addr); | |
342cd0ab CD |
280 | |
281 | if (pud_none_or_clear_bad(pud)) { | |
6060df84 | 282 | pmd = pmd_alloc_one(NULL, addr); |
342cd0ab CD |
283 | if (!pmd) { |
284 | kvm_err("Cannot allocate Hyp pmd\n"); | |
285 | err = -ENOMEM; | |
286 | goto out; | |
287 | } | |
288 | pud_populate(NULL, pud, pmd); | |
4f728276 | 289 | get_page(virt_to_page(pud)); |
5a677ce0 | 290 | kvm_flush_dcache_to_poc(pud, sizeof(*pud)); |
342cd0ab CD |
291 | } |
292 | ||
293 | next = pgd_addr_end(addr, end); | |
6060df84 | 294 | err = create_hyp_pmd_mappings(pud, addr, next, pfn, prot); |
342cd0ab CD |
295 | if (err) |
296 | goto out; | |
6060df84 | 297 | pfn += (next - addr) >> PAGE_SHIFT; |
3562c76d | 298 | } while (addr = next, addr != end); |
342cd0ab CD |
299 | out: |
300 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
301 | return err; | |
302 | } | |
303 | ||
304 | /** | |
06e8c3b0 | 305 | * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode |
342cd0ab CD |
306 | * @from: The virtual kernel start address of the range |
307 | * @to: The virtual kernel end address of the range (exclusive) | |
308 | * | |
06e8c3b0 MZ |
309 | * The same virtual address as the kernel virtual address is also used |
310 | * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying | |
311 | * physical pages. | |
342cd0ab CD |
312 | */ |
313 | int create_hyp_mappings(void *from, void *to) | |
314 | { | |
6060df84 MZ |
315 | unsigned long phys_addr = virt_to_phys(from); |
316 | unsigned long start = KERN_TO_HYP((unsigned long)from); | |
317 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
318 | ||
319 | /* Check for a valid kernel memory mapping */ | |
320 | if (!virt_addr_valid(from) || !virt_addr_valid(to - 1)) | |
321 | return -EINVAL; | |
322 | ||
323 | return __create_hyp_mappings(hyp_pgd, start, end, | |
324 | __phys_to_pfn(phys_addr), PAGE_HYP); | |
342cd0ab CD |
325 | } |
326 | ||
327 | /** | |
06e8c3b0 MZ |
328 | * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode |
329 | * @from: The kernel start VA of the range | |
330 | * @to: The kernel end VA of the range (exclusive) | |
6060df84 | 331 | * @phys_addr: The physical start address which gets mapped |
06e8c3b0 MZ |
332 | * |
333 | * The resulting HYP VA is the same as the kernel VA, modulo | |
334 | * HYP_PAGE_OFFSET. | |
342cd0ab | 335 | */ |
6060df84 | 336 | int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr) |
342cd0ab | 337 | { |
6060df84 MZ |
338 | unsigned long start = KERN_TO_HYP((unsigned long)from); |
339 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
340 | ||
341 | /* Check for a valid kernel IO mapping */ | |
342 | if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1)) | |
343 | return -EINVAL; | |
344 | ||
345 | return __create_hyp_mappings(hyp_pgd, start, end, | |
346 | __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE); | |
342cd0ab CD |
347 | } |
348 | ||
d5d8184d CD |
349 | /** |
350 | * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation. | |
351 | * @kvm: The KVM struct pointer for the VM. | |
352 | * | |
353 | * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can | |
354 | * support either full 40-bit input addresses or limited to 32-bit input | |
355 | * addresses). Clears the allocated pages. | |
356 | * | |
357 | * Note we don't need locking here as this is only called when the VM is | |
358 | * created, which can only be done once. | |
359 | */ | |
360 | int kvm_alloc_stage2_pgd(struct kvm *kvm) | |
361 | { | |
362 | pgd_t *pgd; | |
363 | ||
364 | if (kvm->arch.pgd != NULL) { | |
365 | kvm_err("kvm_arch already initialized?\n"); | |
366 | return -EINVAL; | |
367 | } | |
368 | ||
369 | pgd = (pgd_t *)__get_free_pages(GFP_KERNEL, S2_PGD_ORDER); | |
370 | if (!pgd) | |
371 | return -ENOMEM; | |
372 | ||
373 | /* stage-2 pgd must be aligned to its size */ | |
374 | VM_BUG_ON((unsigned long)pgd & (S2_PGD_SIZE - 1)); | |
375 | ||
376 | memset(pgd, 0, PTRS_PER_S2_PGD * sizeof(pgd_t)); | |
c62ee2b2 | 377 | kvm_clean_pgd(pgd); |
d5d8184d CD |
378 | kvm->arch.pgd = pgd; |
379 | ||
380 | return 0; | |
381 | } | |
382 | ||
d5d8184d CD |
383 | /** |
384 | * unmap_stage2_range -- Clear stage2 page table entries to unmap a range | |
385 | * @kvm: The VM pointer | |
386 | * @start: The intermediate physical base address of the range to unmap | |
387 | * @size: The size of the area to unmap | |
388 | * | |
389 | * Clear a range of stage-2 mappings, lowering the various ref-counts. Must | |
390 | * be called while holding mmu_lock (unless for freeing the stage2 pgd before | |
391 | * destroying the VM), otherwise another faulting VCPU may come in and mess | |
392 | * with things behind our backs. | |
393 | */ | |
394 | static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size) | |
395 | { | |
4f728276 | 396 | unmap_range(kvm->arch.pgd, start, size); |
d5d8184d CD |
397 | } |
398 | ||
399 | /** | |
400 | * kvm_free_stage2_pgd - free all stage-2 tables | |
401 | * @kvm: The KVM struct pointer for the VM. | |
402 | * | |
403 | * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all | |
404 | * underlying level-2 and level-3 tables before freeing the actual level-1 table | |
405 | * and setting the struct pointer to NULL. | |
406 | * | |
407 | * Note we don't need locking here as this is only called when the VM is | |
408 | * destroyed, which can only be done once. | |
409 | */ | |
410 | void kvm_free_stage2_pgd(struct kvm *kvm) | |
411 | { | |
412 | if (kvm->arch.pgd == NULL) | |
413 | return; | |
414 | ||
415 | unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE); | |
416 | free_pages((unsigned long)kvm->arch.pgd, S2_PGD_ORDER); | |
417 | kvm->arch.pgd = NULL; | |
418 | } | |
419 | ||
420 | ||
421 | static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
422 | phys_addr_t addr, const pte_t *new_pte, bool iomap) | |
423 | { | |
424 | pgd_t *pgd; | |
425 | pud_t *pud; | |
426 | pmd_t *pmd; | |
427 | pte_t *pte, old_pte; | |
428 | ||
429 | /* Create 2nd stage page table mapping - Level 1 */ | |
430 | pgd = kvm->arch.pgd + pgd_index(addr); | |
431 | pud = pud_offset(pgd, addr); | |
432 | if (pud_none(*pud)) { | |
433 | if (!cache) | |
434 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
435 | pmd = mmu_memory_cache_alloc(cache); | |
436 | pud_populate(NULL, pud, pmd); | |
d5d8184d | 437 | get_page(virt_to_page(pud)); |
c62ee2b2 MZ |
438 | } |
439 | ||
440 | pmd = pmd_offset(pud, addr); | |
d5d8184d CD |
441 | |
442 | /* Create 2nd stage page table mapping - Level 2 */ | |
443 | if (pmd_none(*pmd)) { | |
444 | if (!cache) | |
445 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
446 | pte = mmu_memory_cache_alloc(cache); | |
c62ee2b2 | 447 | kvm_clean_pte(pte); |
d5d8184d | 448 | pmd_populate_kernel(NULL, pmd, pte); |
d5d8184d | 449 | get_page(virt_to_page(pmd)); |
c62ee2b2 MZ |
450 | } |
451 | ||
452 | pte = pte_offset_kernel(pmd, addr); | |
d5d8184d CD |
453 | |
454 | if (iomap && pte_present(*pte)) | |
455 | return -EFAULT; | |
456 | ||
457 | /* Create 2nd stage page table mapping - Level 3 */ | |
458 | old_pte = *pte; | |
459 | kvm_set_pte(pte, *new_pte); | |
460 | if (pte_present(old_pte)) | |
48762767 | 461 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
d5d8184d CD |
462 | else |
463 | get_page(virt_to_page(pte)); | |
464 | ||
465 | return 0; | |
466 | } | |
467 | ||
468 | /** | |
469 | * kvm_phys_addr_ioremap - map a device range to guest IPA | |
470 | * | |
471 | * @kvm: The KVM pointer | |
472 | * @guest_ipa: The IPA at which to insert the mapping | |
473 | * @pa: The physical address of the device | |
474 | * @size: The size of the mapping | |
475 | */ | |
476 | int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa, | |
477 | phys_addr_t pa, unsigned long size) | |
478 | { | |
479 | phys_addr_t addr, end; | |
480 | int ret = 0; | |
481 | unsigned long pfn; | |
482 | struct kvm_mmu_memory_cache cache = { 0, }; | |
483 | ||
484 | end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK; | |
485 | pfn = __phys_to_pfn(pa); | |
486 | ||
487 | for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) { | |
c62ee2b2 MZ |
488 | pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE); |
489 | kvm_set_s2pte_writable(&pte); | |
d5d8184d CD |
490 | |
491 | ret = mmu_topup_memory_cache(&cache, 2, 2); | |
492 | if (ret) | |
493 | goto out; | |
494 | spin_lock(&kvm->mmu_lock); | |
495 | ret = stage2_set_pte(kvm, &cache, addr, &pte, true); | |
496 | spin_unlock(&kvm->mmu_lock); | |
497 | if (ret) | |
498 | goto out; | |
499 | ||
500 | pfn++; | |
501 | } | |
502 | ||
503 | out: | |
504 | mmu_free_memory_cache(&cache); | |
505 | return ret; | |
506 | } | |
507 | ||
94f8e641 CD |
508 | static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa, |
509 | gfn_t gfn, struct kvm_memory_slot *memslot, | |
510 | unsigned long fault_status) | |
511 | { | |
512 | pte_t new_pte; | |
513 | pfn_t pfn; | |
514 | int ret; | |
515 | bool write_fault, writable; | |
516 | unsigned long mmu_seq; | |
517 | struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache; | |
518 | ||
7393b599 | 519 | write_fault = kvm_is_write_fault(kvm_vcpu_get_hsr(vcpu)); |
94f8e641 CD |
520 | if (fault_status == FSC_PERM && !write_fault) { |
521 | kvm_err("Unexpected L2 read permission error\n"); | |
522 | return -EFAULT; | |
523 | } | |
524 | ||
525 | /* We need minimum second+third level pages */ | |
526 | ret = mmu_topup_memory_cache(memcache, 2, KVM_NR_MEM_OBJS); | |
527 | if (ret) | |
528 | return ret; | |
529 | ||
530 | mmu_seq = vcpu->kvm->mmu_notifier_seq; | |
531 | /* | |
532 | * Ensure the read of mmu_notifier_seq happens before we call | |
533 | * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk | |
534 | * the page we just got a reference to gets unmapped before we have a | |
535 | * chance to grab the mmu_lock, which ensure that if the page gets | |
536 | * unmapped afterwards, the call to kvm_unmap_hva will take it away | |
537 | * from us again properly. This smp_rmb() interacts with the smp_wmb() | |
538 | * in kvm_mmu_notifier_invalidate_<page|range_end>. | |
539 | */ | |
540 | smp_rmb(); | |
541 | ||
542 | pfn = gfn_to_pfn_prot(vcpu->kvm, gfn, write_fault, &writable); | |
543 | if (is_error_pfn(pfn)) | |
544 | return -EFAULT; | |
545 | ||
546 | new_pte = pfn_pte(pfn, PAGE_S2); | |
547 | coherent_icache_guest_page(vcpu->kvm, gfn); | |
548 | ||
549 | spin_lock(&vcpu->kvm->mmu_lock); | |
550 | if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) | |
551 | goto out_unlock; | |
552 | if (writable) { | |
c62ee2b2 | 553 | kvm_set_s2pte_writable(&new_pte); |
94f8e641 CD |
554 | kvm_set_pfn_dirty(pfn); |
555 | } | |
556 | stage2_set_pte(vcpu->kvm, memcache, fault_ipa, &new_pte, false); | |
557 | ||
558 | out_unlock: | |
559 | spin_unlock(&vcpu->kvm->mmu_lock); | |
560 | kvm_release_pfn_clean(pfn); | |
561 | return 0; | |
562 | } | |
563 | ||
564 | /** | |
565 | * kvm_handle_guest_abort - handles all 2nd stage aborts | |
566 | * @vcpu: the VCPU pointer | |
567 | * @run: the kvm_run structure | |
568 | * | |
569 | * Any abort that gets to the host is almost guaranteed to be caused by a | |
570 | * missing second stage translation table entry, which can mean that either the | |
571 | * guest simply needs more memory and we must allocate an appropriate page or it | |
572 | * can mean that the guest tried to access I/O memory, which is emulated by user | |
573 | * space. The distinction is based on the IPA causing the fault and whether this | |
574 | * memory region has been registered as standard RAM by user space. | |
575 | */ | |
342cd0ab CD |
576 | int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run) |
577 | { | |
94f8e641 CD |
578 | unsigned long fault_status; |
579 | phys_addr_t fault_ipa; | |
580 | struct kvm_memory_slot *memslot; | |
581 | bool is_iabt; | |
582 | gfn_t gfn; | |
583 | int ret, idx; | |
584 | ||
52d1dba9 | 585 | is_iabt = kvm_vcpu_trap_is_iabt(vcpu); |
7393b599 | 586 | fault_ipa = kvm_vcpu_get_fault_ipa(vcpu); |
94f8e641 | 587 | |
7393b599 MZ |
588 | trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu), |
589 | kvm_vcpu_get_hfar(vcpu), fault_ipa); | |
94f8e641 CD |
590 | |
591 | /* Check the stage-2 fault is trans. fault or write fault */ | |
1cc287dd | 592 | fault_status = kvm_vcpu_trap_get_fault(vcpu); |
94f8e641 | 593 | if (fault_status != FSC_FAULT && fault_status != FSC_PERM) { |
52d1dba9 MZ |
594 | kvm_err("Unsupported fault status: EC=%#x DFCS=%#lx\n", |
595 | kvm_vcpu_trap_get_class(vcpu), fault_status); | |
94f8e641 CD |
596 | return -EFAULT; |
597 | } | |
598 | ||
599 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
600 | ||
601 | gfn = fault_ipa >> PAGE_SHIFT; | |
602 | if (!kvm_is_visible_gfn(vcpu->kvm, gfn)) { | |
603 | if (is_iabt) { | |
604 | /* Prefetch Abort on I/O address */ | |
7393b599 | 605 | kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu)); |
94f8e641 CD |
606 | ret = 1; |
607 | goto out_unlock; | |
608 | } | |
609 | ||
610 | if (fault_status != FSC_FAULT) { | |
611 | kvm_err("Unsupported fault status on io memory: %#lx\n", | |
612 | fault_status); | |
613 | ret = -EFAULT; | |
614 | goto out_unlock; | |
615 | } | |
616 | ||
cfe3950c MZ |
617 | /* |
618 | * The IPA is reported as [MAX:12], so we need to | |
619 | * complement it with the bottom 12 bits from the | |
620 | * faulting VA. This is always 12 bits, irrespective | |
621 | * of the page size. | |
622 | */ | |
623 | fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1); | |
45e96ea6 | 624 | ret = io_mem_abort(vcpu, run, fault_ipa); |
94f8e641 CD |
625 | goto out_unlock; |
626 | } | |
627 | ||
628 | memslot = gfn_to_memslot(vcpu->kvm, gfn); | |
94f8e641 CD |
629 | |
630 | ret = user_mem_abort(vcpu, fault_ipa, gfn, memslot, fault_status); | |
631 | if (ret == 0) | |
632 | ret = 1; | |
633 | out_unlock: | |
634 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
635 | return ret; | |
342cd0ab CD |
636 | } |
637 | ||
d5d8184d CD |
638 | static void handle_hva_to_gpa(struct kvm *kvm, |
639 | unsigned long start, | |
640 | unsigned long end, | |
641 | void (*handler)(struct kvm *kvm, | |
642 | gpa_t gpa, void *data), | |
643 | void *data) | |
644 | { | |
645 | struct kvm_memslots *slots; | |
646 | struct kvm_memory_slot *memslot; | |
647 | ||
648 | slots = kvm_memslots(kvm); | |
649 | ||
650 | /* we only care about the pages that the guest sees */ | |
651 | kvm_for_each_memslot(memslot, slots) { | |
652 | unsigned long hva_start, hva_end; | |
653 | gfn_t gfn, gfn_end; | |
654 | ||
655 | hva_start = max(start, memslot->userspace_addr); | |
656 | hva_end = min(end, memslot->userspace_addr + | |
657 | (memslot->npages << PAGE_SHIFT)); | |
658 | if (hva_start >= hva_end) | |
659 | continue; | |
660 | ||
661 | /* | |
662 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
663 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
664 | */ | |
665 | gfn = hva_to_gfn_memslot(hva_start, memslot); | |
666 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
667 | ||
668 | for (; gfn < gfn_end; ++gfn) { | |
669 | gpa_t gpa = gfn << PAGE_SHIFT; | |
670 | handler(kvm, gpa, data); | |
671 | } | |
672 | } | |
673 | } | |
674 | ||
675 | static void kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
676 | { | |
677 | unmap_stage2_range(kvm, gpa, PAGE_SIZE); | |
48762767 | 678 | kvm_tlb_flush_vmid_ipa(kvm, gpa); |
d5d8184d CD |
679 | } |
680 | ||
681 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
682 | { | |
683 | unsigned long end = hva + PAGE_SIZE; | |
684 | ||
685 | if (!kvm->arch.pgd) | |
686 | return 0; | |
687 | ||
688 | trace_kvm_unmap_hva(hva); | |
689 | handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL); | |
690 | return 0; | |
691 | } | |
692 | ||
693 | int kvm_unmap_hva_range(struct kvm *kvm, | |
694 | unsigned long start, unsigned long end) | |
695 | { | |
696 | if (!kvm->arch.pgd) | |
697 | return 0; | |
698 | ||
699 | trace_kvm_unmap_hva_range(start, end); | |
700 | handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL); | |
701 | return 0; | |
702 | } | |
703 | ||
704 | static void kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
705 | { | |
706 | pte_t *pte = (pte_t *)data; | |
707 | ||
708 | stage2_set_pte(kvm, NULL, gpa, pte, false); | |
709 | } | |
710 | ||
711 | ||
712 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
713 | { | |
714 | unsigned long end = hva + PAGE_SIZE; | |
715 | pte_t stage2_pte; | |
716 | ||
717 | if (!kvm->arch.pgd) | |
718 | return; | |
719 | ||
720 | trace_kvm_set_spte_hva(hva); | |
721 | stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2); | |
722 | handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte); | |
723 | } | |
724 | ||
725 | void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
726 | { | |
727 | mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); | |
728 | } | |
729 | ||
342cd0ab CD |
730 | phys_addr_t kvm_mmu_get_httbr(void) |
731 | { | |
342cd0ab CD |
732 | return virt_to_phys(hyp_pgd); |
733 | } | |
734 | ||
5a677ce0 MZ |
735 | phys_addr_t kvm_mmu_get_boot_httbr(void) |
736 | { | |
737 | return virt_to_phys(boot_hyp_pgd); | |
738 | } | |
739 | ||
740 | phys_addr_t kvm_get_idmap_vector(void) | |
741 | { | |
742 | return hyp_idmap_vector; | |
743 | } | |
744 | ||
342cd0ab CD |
745 | int kvm_mmu_init(void) |
746 | { | |
2fb41059 MZ |
747 | int err; |
748 | ||
5a677ce0 MZ |
749 | hyp_idmap_start = virt_to_phys(__hyp_idmap_text_start); |
750 | hyp_idmap_end = virt_to_phys(__hyp_idmap_text_end); | |
751 | hyp_idmap_vector = virt_to_phys(__kvm_hyp_init); | |
752 | ||
753 | if ((hyp_idmap_start ^ hyp_idmap_end) & PAGE_MASK) { | |
754 | /* | |
755 | * Our init code is crossing a page boundary. Allocate | |
756 | * a bounce page, copy the code over and use that. | |
757 | */ | |
758 | size_t len = __hyp_idmap_text_end - __hyp_idmap_text_start; | |
759 | phys_addr_t phys_base; | |
760 | ||
761 | init_bounce_page = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
762 | if (!init_bounce_page) { | |
763 | kvm_err("Couldn't allocate HYP init bounce page\n"); | |
764 | err = -ENOMEM; | |
765 | goto out; | |
766 | } | |
767 | ||
768 | memcpy(init_bounce_page, __hyp_idmap_text_start, len); | |
769 | /* | |
770 | * Warning: the code we just copied to the bounce page | |
771 | * must be flushed to the point of coherency. | |
772 | * Otherwise, the data may be sitting in L2, and HYP | |
773 | * mode won't be able to observe it as it runs with | |
774 | * caches off at that point. | |
775 | */ | |
776 | kvm_flush_dcache_to_poc(init_bounce_page, len); | |
777 | ||
778 | phys_base = virt_to_phys(init_bounce_page); | |
779 | hyp_idmap_vector += phys_base - hyp_idmap_start; | |
780 | hyp_idmap_start = phys_base; | |
781 | hyp_idmap_end = phys_base + len; | |
782 | ||
783 | kvm_info("Using HYP init bounce page @%lx\n", | |
784 | (unsigned long)phys_base); | |
785 | } | |
786 | ||
2fb41059 | 787 | hyp_pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL); |
5a677ce0 MZ |
788 | boot_hyp_pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL); |
789 | if (!hyp_pgd || !boot_hyp_pgd) { | |
d5d8184d | 790 | kvm_err("Hyp mode PGD not allocated\n"); |
2fb41059 MZ |
791 | err = -ENOMEM; |
792 | goto out; | |
793 | } | |
794 | ||
795 | /* Create the idmap in the boot page tables */ | |
796 | err = __create_hyp_mappings(boot_hyp_pgd, | |
797 | hyp_idmap_start, hyp_idmap_end, | |
798 | __phys_to_pfn(hyp_idmap_start), | |
799 | PAGE_HYP); | |
800 | ||
801 | if (err) { | |
802 | kvm_err("Failed to idmap %lx-%lx\n", | |
803 | hyp_idmap_start, hyp_idmap_end); | |
804 | goto out; | |
d5d8184d CD |
805 | } |
806 | ||
5a677ce0 MZ |
807 | /* Map the very same page at the trampoline VA */ |
808 | err = __create_hyp_mappings(boot_hyp_pgd, | |
809 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
810 | __phys_to_pfn(hyp_idmap_start), | |
811 | PAGE_HYP); | |
812 | if (err) { | |
813 | kvm_err("Failed to map trampoline @%lx into boot HYP pgd\n", | |
814 | TRAMPOLINE_VA); | |
815 | goto out; | |
816 | } | |
817 | ||
818 | /* Map the same page again into the runtime page tables */ | |
819 | err = __create_hyp_mappings(hyp_pgd, | |
820 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
821 | __phys_to_pfn(hyp_idmap_start), | |
822 | PAGE_HYP); | |
823 | if (err) { | |
824 | kvm_err("Failed to map trampoline @%lx into runtime HYP pgd\n", | |
825 | TRAMPOLINE_VA); | |
826 | goto out; | |
827 | } | |
828 | ||
d5d8184d | 829 | return 0; |
2fb41059 | 830 | out: |
4f728276 | 831 | free_hyp_pgds(); |
2fb41059 | 832 | return err; |
342cd0ab | 833 | } |